Crypto-DSEDA: A Domain-specific EDA Flow for CiM-based Cryptographic Accelerators

Cryptographic algorithms are infrastructures of information security. They have high requirements in performance and energy consumption, but they suffer from the "memory wall" bottleneck on traditional computers. Computing-in-memory (CiM) is considered a promising technology to accelerate cryptographic algorithms. Since traditional EDA tools are not suitable for exploring CiM architectures, existing cryptographic accelerators involve a large amount of manual design efforts. This work proposes Crypto-DSEDA, an auto-generation framework for exploring CiM-based cryptographic accelerators. Crypto-DSEDA includes an automatic architecture generation pipeline that takes behavioral-level descriptions as input, optimizes dataflow and resource allocation through design space exploration, and finally generates the accelerator architecture and dataflow of the target algorithm. We evaluate Crypto-DSEDA on popular cryptographic algorithms, revealing that the auto-generated architectures that perform dataflow graph optimizations achieve significant improvements in latency and power consumption. Compared with a CiM-based implementation, Crypto-DSEDA also shows significant performance and energy efficiency improvements.